Validation of empirical formulas to derive model parameters for sands

Validation of empirical formulas to derive model parameters for sands R.B.J. Brinkgreve Geo-Engineering Section, Delft University of Technology, Delft, Netherlands/Plaxis B.V., Delft, Netherlands E. Engin Plaxis B.V., Delft, Netherlands H.K. Engin Geo-Engineering Section, Delft University of Technology, Delft, Netherlands ABSTRACT: The right selection of soil model parameters is essential to make good predictions in geoengineering projects where the Finite Element Method (FEM) is used. In order to support geotechnical engineers with soil model parameter selection, empirical formulas have been developed to derive the model parameters of the Plaxis Hardening Soil model with small-strain stiffness (HSsmall), on the basis of a characteristic property (relative density for sands and plasticity index for clays). This paper shows a validation of formulas for sands which have been derived from published soil testing data. The main goal of the empirical formulas is to give a reasonable first order approximation of soil behaviour in FEM calculations, covering a wide range of sands. In a case study it is demonstrated that the empirical formulas work reasonably well to get a first estimate of deformations and stress developments for a real project. INTRODUCTION In the last decades, many researchers have investigated the properties of sands and clays, and they have published formulas, charts and tables as a general erence to support the design of geotechnical structures (e.g. Kulhawy & Mayne (99)). Most general data ers to soil strength properties, such as a friction angle for sand or undrained shear strength for clay, which can be primarily used for stability analysis and ultimate limit state design (ULS). In contrast to ULS, serviceability state design (SLS) requires stiffness properties to be known. Several researchers have published correlations between stiffness and strength, index properties and/or state parameters. For sands, many correlations exist with the relative density, whereas for clay many correlations exist with the plasticity index. Over the last twenty years, the Finite Element Method (FEM) has gained much popularity for geoengineering and design. In FEM, the mechanical behaviour of soils is simulated by means of constitutive models, in which model parameters are used to quantify particular features of soil behaviour. Constitutive models may range from simple to very advanced models. In general, simple models require only a limited number of parameters to be selected, but they may also lack some essential features of soil behaviour. More advanced models may include more features of soil behaviour, but also require more parameters to be selected on the basis of soil investigation data. The latter often discourages FEM users to use advanced models for daily geoengineering projects. Some authors have published data sets with predefined model parameters for particular types of soils (e.g. Duncan et al. (98) containing data sets for the Duncan-Chang model). It is challenging but probably unrealistic to extend this idea and cover all existing soil types on a world-wide scale. As an alternative, the authors of this paper are developing empirical formulas to derive all parameters of the Plaxis HSsmall model on the basis of very limited geotechnical data in an attempt to stimulate the use of advanced models for soil in general. This seems contradictive, but since most soil properties are sort of correlated, it is believed that a first approximation could be obtained with reasonable accuracy. The authors are convinced that such a first approximation using an advanced model gives better results than a first approximation with a simple model. It is definitely not the idea to abandon detailed site-specific soil investigation, but the use of the formulas can be very helpful, especially in the early design stage of a project when very limited soil data from the site are available. In the next chapter formulas are presented for sand to derive the model parameters of the HSsmall model on the basis of the relative density. Even if the relative density is not precisely known, it could be estimated on the basis of very preliminary soil data. The formulas have been derived by regression analysis on a collection of soil data (general soil data, triaxial test data, oedometer test data, etc.) from

Jeffries & Been (6) and others. The third chapter describes a validation of the formulas by comparing the results of model simulations with real lab test data for different types of sands. Chapter four describes a benchmark example (Schweiger, and ), in which the formulas have been applied to Berlin sand to predict deformations and structural forces due to the excavation. Finally, some conclusions are drawn. FORMULAS FOR SAND The relative density (RD) is defined as (e max e) / (e max e min ), where e is the current void ratio, e max is the maximum void ratio (loosest packing) and e min is the minimum void ratio (densest packing). The relative density is usually presented as a percentage, as also used in this paper. Before considering the parameters of the HSsmall model, the relative density can already be used to estimate the unit weights of sand for practical applications by means of the following formulas: γ unsat = +. RD / [ kn / m ] () γ sat = 9 +.6 RD / [ kn / m ] () The HSsmall model contains four different stiffness parameters, each of them quantifying the erence stiffness in a particular stress path for a given erence stress level, p. For a detailed description of the HSsmall model and the meaning of its parameters, erence is made to Benz (7) and Brinkgreve et al. (8). For (quartz) sand, stiffness is supposed to vary linearly with RD. The following formulas are proposed for the erence stiffness parameters, considering p = kn/m : Go (MPa) 8 6 Toyoura Sand- (subangular) Toyoura Sand- (subangular) Toyoura Sand - (subangular) Toyoura Sand - (subangular) Ham River Sand (subangular) Ham River Sand (subangular) Hostun Sand Ticino Sand Ticino Sand (subangular) Ticino Sand (subangular) According to Eqn (6) E = RD kn m () 6 / [ / ] E RD kn m oed ur = 6 / [ / ] () E RD kn m = 8 / [ / ] () G = + RD kn m (6) 6 68 / [ / ] Figure shows the variation of G with RD for different sands, in comparison with Equation 6. The actual stiffness is stress-dependent. The rate of stress dependency, m, is observed to be negatively correlated with the density. The following formula is proposed for m: m =.7 RD / [ ] (7) Poisson s ratio for unloading and reloading, ν ur, is taken.. The parameter relating the modulus reduction curve to the cyclic shear strain level is γ.7, for which the following formula is proposed: γ = (8).7 ( RD /) [ ] The following formulas are proposed for the strength-related properties: ϕ ' = 8 +. RD / [ ] (9) ψ = +. RD / [ ] () R f = RD / 8 [ ] () These values should be used for drained conditions. Table gives an example of parameter values for loose, medium, dense and very dense sand using the above formulas. 8 6...6.8 Relative Density, RD Figure. Comparison of formula for small-strain stiffness for different sands at different densities

Saturated unit weight, γsat (kn/m ) 9 Toyoura Sand Ticino Sand Nerlerk Berm Alaskan Sand Ersak Sand Equation () 8 6 8 Relative density, RD (%) Figure. Comparison of formula for saturated unit weight for different sands at different densities Table. Examples of model parameters for sands with different relative densities RD γ unsat γ sat E E oed E ur G m - kn/m kn/m kn/m kn/m kn/m - 6. 9. 77.6 7. 9.8 9 9. 8 8.. 8 8. 9..6 6 6 8 8.88 RD γ.7 ϕ ψ R f - - -.8 -...969. -...98 8. - 8. 8..9. -...87 VALIDATION OF FORMULAS The formulas for unit weight have been validated for different sands at different densities with data from Jeffries & Been (6). Both the saturated unit weight γ sat and the dry unit weight γ dry were reported. Figure shows the variation of γ sat with RD for different sands, in comparison with Equation. The unsaturated unit weight γ unsat, as proposed in formula, is a realistic practical value in between γ dry and γ sat. To validate the formulas for the HSsmall stiffness parameters, a comparison has been made between data from real drained triaxial tests on different types of sand with the results from numerical simulations with the HSsmall model. Figure shows the results for triaxial tests on Karlsruhe sand of different densities at kn/m cell pressure, based on data by Wu (99). Another series of drained triaxial tests have been analysed for loose and very dense Sacramento River sand at different cell pressures, based on data by Lee (96). Some of these results are shown in Figure. Stress Ratio ( σ / σ ) Volumetric Strain, ε v (%) - - Axial Strain, ε (%) 6 8 RD = 7.9% RD* = 7.9 % RD = 66.% RD* = 66. % RD = 9.% RD* = 9.% Figure. Comparison of drained triaxial tests on Karlsruhe sand with HSsmall model for different relative densities (* indicates experimental data, after Wu, 99) Another comparison has been made for dense Hokksund sand, based on a drained triaxial test and an oedometer test as reported by Yang (). The results are shown in Figures and 6.

Stress Ratio, σ₁/σ₃ Volumetric Strain, ε v (%) - - - - - -6 Axial Strain ε₁ (%) σ = 98 kn/m² *σ = 98 kn/m² σ = 96 kn/m² *σ = 96 kn/m² σ = kn/m² *σ = kn/m² σ = kn/m² *σ = kn/m² Figure. Comparison of drained triaxial tests on loose Sacramento river sand (RD=8%) with HSsmall model for different cell pressures (* indicates experimental data, after Lee, 96) From the different test results the following can be concluded: In most cases, the formulas show a (small) overestimation of the stiffness in triaxial loading. Strength and dilatancy are generally underestimated. High friction and dilatancy may reduce in reality as a result of shearing and softening, which is not included in the HSsmall model. Hence, a small under-estimation of the peak strength might even be desirable. The stiffness in oedometer loading is difficult to match with soil testing data. The data presented in Figure 6 seems reasonable, but other tests were less successful. Reason for this is that in oedometer tests the stress range is usually quite large and the density changes significantly, so one RDvalue cannot cover the full test. Hostun sand seems to be rather soft. The formulas tend to significantly over-estimate the stiffness of Hostun sand (not presented herein). It should be noted that these tests primarily validate the formulas for loading stiffnesses, friction and dilatancy (Equations,, 9 and ) and to a lesser extend the formulas for the other model parameters, but some of these will be considered in the next chapter. σ₁ - σ₃ Volumetric Strain, ε v (%) Figure. Comparison of drained triaxial test on dense Hokksund sand with HSsmall model Vertical Stress,σ₁' 8 6 8 6-9 6 RD = 8% - Plaxis HSsmall RD* 8% (Test)...8..6 Axial Strain, ε₁ Figure 6. Comparison of oedometer loading and unloading test on dense Hokksund sand with HSsmall model CASE STUDY RD = 98% - Plaxis HSsmall RD* = 98% (Test) Axial Strain, ε₁(%) In this chapter a case study is presented, based on the benchmark example as described by Schweiger (). It concerns an excavation in Berlin sand, supported by a triple anchored retaining wall. The excavation is 6.8m deep and the wall is m long. Three rows of anchors are used, starting just above the intermediate excavation levels of.8m, 9.m and.m depth. The wall has been modelled by Mindlin beam elements; the wall-soil interaction by interface elements; the anchors by a combination of membrane elements (grouted body) and two-node spring elements (anchor rod). Figure 7 shows the used D finite element mesh composed of -node (cubic strain) elements, and some model details. The soil was reported to be medium dense Berlin sand (from an undisturbed sample at 8 m depth). Although more soil data was provided, only the information medium dense was used here and inter-

- 6 8 - - -6-8 - Anchor Prestress Distance Area Σ Length Row kn m cm m 768. 9.8 9.. 98..8 Figure 7. Geometry and mesh of triple anchored excavation in Berlin sand (with details from Schweiger, ) preted as RD = %, at least for the upper m of sand. The next m was considered to be denser, with an assumed RD of 8%, whereas the lower 6 m was assumed to be very dense with an assumed RD of %. The formulas presented in Chapter were used to estimate the model parameters (see Table for RD = %, 8% and % respectively). In this case study the unloading stiffness and smallstrain stiffness parameters are more relevant than the loading stiffnesses. The interface strength was related to the strength in the surrounding soil, and reduced by a factor.8, as in the original benchmark. The structural properties were taken from Schweiger (), i.e. E steel =. 8 kn/m and E concrete =. 7 kn/m. The excavation process was simulated in 8 different stages, starting with the installation of the wall, followed by the four excavation stages (including lowering of the water table), and in between separate stages to install and pre-stress the next anchor row. Horizontal displacements (mm) -6 - - Reference Solution (Schweiger, ) Solution with proposed formulas Figure 8. Comparison of horizontal displacement profile of the wall (after Schweiger, ) Depth below ground level (m) Figure 7 shows the horizontal displacements of the wall after full excavation, and a comparison with the erence solution by Schweiger (). The calculated maximum displacement of the wall is mm. This is about. times higher than the (corrected) measurements. The overall deformation shape is quite similar, but shows a mm shift compared to the erence solution. This indicates that the soil is not stiff enough, which could indicate that the small-strain stiffness is too low. The distribution of bending moments is quite similar to the erence solution; the maximum value (7 knm/m) is almost equal. Anchor forces are in the right order, but show more variations in the phases after installation than in the erence solution, due to the lower soil stiffness. Considering that the erence solution is based on more detailed soil data, the results presented herein are quite reasonable for a first approximation. Bending moment (kn-m/m) - - Reference Solution (Schweiger, ) Using Proposed Formulas Figure 9. Comparison of bending moment diagrams of the wall with triple anchors (after Schweiger, ) Depth below ground level (m)

CONCLUSIONS Empirical formulas have been presented for sands to select all model parameters of the Plaxis Hardening Soil model with small-strain stiffness, on the basis of the relative density. The formulas have been validated against lab test data for different sands at different densities and different pressures. Moreover, the formulas have been used in a benchmark example involving a triple anchored excavation in Berlin sand. Although not all formulas or model parameters have been validated in sufficient detail, it can be concluded from the current results that the formulas may give a reasonable first approximation of the drained behaviour of (quartz) sands in geoengineering applications. Some formulas may be reconsidered or improved, but it should be realized that a high accuracy can never be achieved unless additional information (such as grain size distribution, grain shape, etc.) is taken into account. Nevertheless, the formulas can be quite useful in the beginning of a project when only limited soil data is available. By using general formulas for model parameter selection it is not the idea to abandon detailed soil investigation. Since the formulas cannot provide sufficient accuracy for a final design, more detailed soil investigation remains definitely required. A first analysis based on these formulas may actually help to define a detailed soil investigation plan, because it can give insight in dominant stress paths and critical locations in the project. Schweiger, H.F. (), Ergebnisse des Berechnungsbeispieles Nr. -fach verankerte Baugrube. Tagungsband Workshop Verformungsprognose für tiefe Baugruben, Stuttgart, 7-67 (in German). Schweiger, H.F. (), Results from numerical benchmark exercises in geotechnics, Proceedings of th European Conference on Numerical Methods in Geotechnical Engineering, Paris. Wu, W. (99), The behaviour of very loose sand in the triaxial compression test: Discussion. Canadian Geotechnical Journal, Vol. 7, 9-6. Yang, S.L. () Characterization of the properties of sand-silt mixtures, PhD thesis, Norwegian University of Science and Technology, Norway. REFERENCES Benz, T. (7), Small strain stiffness of soils and its numerical consequences, PhD thesis, University of Stuttgart, Germany. Brinkgreve, R.B.J., Broere, W., Waterman, D. (8), Plaxis D version 9., Material Models Manual. Plaxis BV, Delft. Duncan, J.M., Byrne, P., Wong, K.S. Mabry, P. (98), Geotechnical Engineering Strength, stress-strain and bulk modulus parameters for finite element analyses of stresses and movements in soil masses. Virginia Tech, Dept. of Civil Engineering, Blacksburg. Jefferies, M., Been, K. (6). Soil Liquefaction: A Critical State Approach. Taylor & Francis, Abingdon, UK. Kulhawy, F.H., Mayne, P.W. (99), Manual on estimating soil properties for foundation design, Electric Power Research Institute, California. Lee, K.L. (96), Triaxial compressive strength on saturated sands under cyclic loading conditions, PhD thesis, University of California at Berkeley.